Arkandia Skills

arkandia-skills

*The AI can't read your mind. It reads files.*

A collection of AI coding-agent skills from Arkandia for an agentic SDLC. They follow one arc: give the repository the **context** an agent needs to decide, install the **instrumentation** it cannot ship past, then **deliver** tickets on top of both.

The instrumentation layer splits in two, as the Arkandia Method does. `instrument-project-dotnet` covers the **deterministic** half — what a machine verifies on its own, with no ambiguity: the build, style, secrets, architecture tests, CI. `instrument-agent-dotnet` covers the **non-deterministic** half — the tools the agent may reach and the limits it works inside, as MCP servers and hooks.

Works with Claude Code, OpenCode, Codex, Cursor, and the other agents supported by [`skills.sh`](https://skills.sh).

**[Versión en español →](./README-es.md)**

Why

Coding agents behave well when the repository tells them what they need to know, and badly when critical context lives in someone's head, in Slack, or in an unread Google Doc. That is the first problem these skills solve.

The second is newer: agents now write code faster than a team can review it. Documentation alone does not scale against that — a rule an agent can read is a suggestion, a rule that fails the build is a rule. So verification has to become mechanical, and human review has to be reserved for what machines genuinely cannot judge: intent, design, trade-offs, product alignment.

The design is informed by OpenAI's harness-engineering writing (*"AGENTS.md is a table of contents, not an encyclopedia"*, *"the repository is the system of record"*) and by the **Arkandia Method** framework from the *AI-Driven Development* workshop.

Skills

Skill	What it does	Docs
agent-context-dotnet	Context — bootstraps AGENTS.md, architecture, ADRs, data model, infrastructure and a docs/dotnet.md deep-dive for a .NET repo, then validates the load-bearing claims with